PASCAL - Pattern Analysis, Statistical Modelling and Computational Learning

Minimal penalties for Gaussian model selection.
Pascal Massart and L Birge
Probability Th. and Related Fields pp. 35-35, 2004.

Abstract

This paper is mainly devoted to a precise analysis of what kind of penalties should be used in order to perform model selection via the minimzation of a penalized least squares criterion within some general Gaussian framework.

EPrint Type:Article
Project Keyword:Project Keyword UNSPECIFIED
Subjects:Computational, Information-Theoretic Learning with Statistics
ID Code:631
Deposited By:Michele Sebag
Deposited On:29 December 2004